Chatbot(Ques_Ans).ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "Chatbot(Ques/Ans).ipynb",
      "provenance": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "13tzIq4ZUHaw",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Library Imports\n",
        "import pickle\n",
        "import numpy as np"
      ],
      "execution_count": 462,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "as0STcckUf4R",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#retrieve training data\n",
        "with open('/content/train_qa.txt', 'rb') as f:\n",
        "    train_data = pickle.load(f)"
      ],
      "execution_count": 463,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BSjOYfwRU4iT",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#retrieve test data\n",
        "with open('/content/test_qa.txt', 'rb') as f:\n",
        "    test_data = pickle.load(f)"
      ],
      "execution_count": 464,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3H6XTmbxXIkf",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "27e65cbc-aa77-4eb2-f39a-321c02b73041"
      },
      "source": [
        "#Number of training instances\n",
        "len(train_data)\n",
        "#Number of test instances\n",
        "len(test_data)"
      ],
      "execution_count": 465,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1000"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 465
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fYvXvpyrXi0y",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 277
        },
        "outputId": "fc5d63f2-751b-438b-b6d2-0f58d44a00f5"
      },
      "source": [
        "#Example of one of the instances\n",
        "train_data[10]"
      ],
      "execution_count": 466,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(['Sandra',\n",
              "  'went',\n",
              "  'back',\n",
              "  'to',\n",
              "  'the',\n",
              "  'hallway',\n",
              "  '.',\n",
              "  'Sandra',\n",
              "  'moved',\n",
              "  'to',\n",
              "  'the',\n",
              "  'office',\n",
              "  '.'],\n",
              " ['Is', 'Sandra', 'in', 'the', 'office', '?'],\n",
              " 'yes')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 466
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "js6U0LgAX3bB",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "c98f39ae-874c-4b9e-abea-1742d2e348cc"
      },
      "source": [
        "' '.join(train_data[10][0]) #story"
      ],
      "execution_count": 467,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic": {
              "type": "string"
            },
            "text/plain": [
              "'Sandra went back to the hallway . Sandra moved to the office .'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 467
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kKoBYc8yYJAk",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "8022f8aa-9e1a-4698-eb07-884002bfcf71"
      },
      "source": [
        "' '.join(train_data[10][1]) #question"
      ],
      "execution_count": 468,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic": {
              "type": "string"
            },
            "text/plain": [
              "'Is Sandra in the office ?'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 468
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4I73rnhscndM",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "8990f9e2-c8ef-424f-f360-2e3238bedd5f"
      },
      "source": [
        "train_data[10][2] #answer"
      ],
      "execution_count": 469,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "application/vnd.google.colaboratory.intrinsic": {
              "type": "string"
            },
            "text/plain": [
              "'yes'"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 469
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RIqQUXddn9e5",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#First we need to create a vocabulary with our data\n",
        "#For this we will use the training data only to \n",
        "vocab = set()\n",
        "for story, question, answer in train_data:\n",
        "    vocab = vocab.union(set(story)) #Set returns unique words in the sentence\n",
        "                                    #Union returns the unique common elements from a two sets\n",
        "    vocab = vocab.union(set(question))"
      ],
      "execution_count": 470,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "um6l1bk3n_sn",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "vocab.add('no')\n",
        "vocab.add('yes')"
      ],
      "execution_count": 471,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vyhqGIQdoHji",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 55
        },
        "outputId": "0f0221b1-9002-4e77-defd-acc3f80d5f26"
      },
      "source": [
        "print(vocab)"
      ],
      "execution_count": 472,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "{'moved', 'bathroom', 'in', 'Is', 'hallway', 'dropped', 'down', 'yes', 'kitchen', 'travelled', 'milk', 'bedroom', 'got', 'up', 'went', 'put', '?', 'office', 'Sandra', 'garden', 'discarded', 'to', 'Mary', 'took', 'football', 'no', 'journeyed', 'John', '.', 'left', 'the', 'Daniel', 'there', 'back', 'apple', 'grabbed', 'picked'}\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4sF46kJYoeKN",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Calculate len and add 1 for Keras placeholder - Placeholders are used to feed in the data to the network. \n",
        "#They need a data type, and have optional shape arguements.\n",
        "#They will be empty at first, and then the data will get fed into the placeholder\n",
        "vocab_len = len(vocab) + 1"
      ],
      "execution_count": 473,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fQxWG4ZYozyF",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "448fadd0-b827-4ec3-a284-272f2a38172a"
      },
      "source": [
        "print(vocab_len)"
      ],
      "execution_count": 474,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "38\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Z6Dpcvvho2DH",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 55
        },
        "outputId": "315af5ec-2dd9-4b25-a7a6-3093856dbc30"
      },
      "source": [
        "print(vocab)"
      ],
      "execution_count": 475,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "{'moved', 'bathroom', 'in', 'Is', 'hallway', 'dropped', 'down', 'yes', 'kitchen', 'travelled', 'milk', 'bedroom', 'got', 'up', 'went', 'put', '?', 'office', 'Sandra', 'garden', 'discarded', 'to', 'Mary', 'took', 'football', 'no', 'journeyed', 'John', '.', 'left', 'the', 'Daniel', 'there', 'back', 'apple', 'grabbed', 'picked'}\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "j_cW2ekpo4Ip",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Now we are going to calculate the longest story and the longest question\n",
        "#We need this for the Keras pad sequences. \n",
        "#Keras training layers expect all of the input to have the same length, so \n",
        "#we need to pad \n",
        "all_data = test_data + train_data"
      ],
      "execution_count": 476,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Y8t83e-joEzV",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "all_story_lens = [len(data[0]) for data in all_data]"
      ],
      "execution_count": 477,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "14IJ_LbtoHIB",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "max_story_len = (max(all_story_lens))"
      ],
      "execution_count": 478,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fArY2xWPoJXX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "max_question_len = max([len(data[1]) for data in all_data])"
      ],
      "execution_count": 479,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sk6ZTlykoX6e",
        "colab_type": "text"
      },
      "source": [
        "### Vectorising the data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "o4mXXS0nodvB",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from keras.preprocessing.sequence import pad_sequences\n",
        "from keras.preprocessing.text import Tokenizer\n"
      ],
      "execution_count": 480,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DUsP1Nnkong8",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Create an instance of the tokenizer object:\n",
        "tokenizer = Tokenizer(filters = [])\n",
        "tokenizer.fit_on_texts(vocab)"
      ],
      "execution_count": 481,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Xr85rpqkowsD",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 55
        },
        "outputId": "3474375e-6b85-4217-d501-5aa02fbb0625"
      },
      "source": [
        "#Dictionary that maps every word in our vocab to an index\n",
        "# It has been automatically lowercased\n",
        "#This tokenizer can give different indexes for different words depending on when we run it\n",
        "print(tokenizer.word_index)"
      ],
      "execution_count": 482,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "{'moved': 1, 'bathroom': 2, 'in': 3, 'is': 4, 'hallway': 5, 'dropped': 6, 'down': 7, 'yes': 8, 'kitchen': 9, 'travelled': 10, 'milk': 11, 'bedroom': 12, 'got': 13, 'up': 14, 'went': 15, 'put': 16, '?': 17, 'office': 18, 'sandra': 19, 'garden': 20, 'discarded': 21, 'to': 22, 'mary': 23, 'took': 24, 'football': 25, 'no': 26, 'journeyed': 27, 'john': 28, '.': 29, 'left': 30, 'the': 31, 'daniel': 32, 'there': 33, 'back': 34, 'apple': 35, 'grabbed': 36, 'picked': 37}\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JEC6ZcLZr4wI",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Tokenize the stories, questions and answers:\n",
        "train_story_text = []\n",
        "train_question_text = []\n",
        "train_answers = []\n",
        "#Separating each of the elements\n",
        "for story,question,answer in train_data:\n",
        "    train_story_text.append(story)\n",
        "    train_question_text.append(question) \n",
        "    train_answers.append(answer)\n",
        "#Coverting the text into the indexes \n",
        "train_story_seq = tokenizer.texts_to_sequences(train_story_text)"
      ],
      "execution_count": 483,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7dPTdDSEoLnI",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def vectorize_stories(data,word_index = tokenizer.word_index, max_story_len = max_story_len, max_question_len = max_question_len):\n",
        "    #vectorized stories:\n",
        "    X = []\n",
        "    #vectorized questions:\n",
        "    Xq = []\n",
        "    #vectorized answers:\n",
        "    Y = []\n",
        "    \n",
        "    for story, question, answer in data:\n",
        "        #Getting indexes for each word in the story\n",
        "        x = [word_index[word.lower()] for word in story]\n",
        "        #Getting indexes for each word in the story\n",
        "        xq = [word_index[word.lower()] for word in question]\n",
        "        #For the answers\n",
        "        y = np.zeros(len(word_index) + 1) #Index 0 Reserved when padding the sequences\n",
        "        y[word_index[answer]] = 1\n",
        "        \n",
        "        X.append(x)\n",
        "        Xq.append(xq)\n",
        "        Y.append(y)\n",
        "        \n",
        "    #Now we have to pad these sequences:\n",
        "    return(pad_sequences(X,maxlen=max_story_len), pad_sequences(Xq, maxlen=max_question_len), np.array(Y))"
      ],
      "execution_count": 484,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WgI8sQjwrU6D",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "inputs_train, questions_train, answers_train = vectorize_stories(train_data)"
      ],
      "execution_count": 485,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RlKkANi1rVoh",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "inputs_test, questions_test, answers_test = vectorize_stories(test_data)"
      ],
      "execution_count": 486,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "u_Z0XmfwrZXp",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 191
        },
        "outputId": "c223ca74-201f-4b96-db2a-a73d54222a6f"
      },
      "source": [
        "inputs_train[0]"
      ],
      "execution_count": 487,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([ 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,\n",
              "        0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,\n",
              "        0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,\n",
              "        0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,\n",
              "        0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,\n",
              "        0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,\n",
              "        0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,\n",
              "        0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,\n",
              "        0,  0,  0,  0,  0,  0,  0,  0, 23,  1, 22, 31,  2, 29, 19, 27, 22,\n",
              "       31, 12, 29], dtype=int32)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 487
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XjovQNI1rb0P",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 225
        },
        "outputId": "064efcea-0ec5-4613-da7e-d4c001188d82"
      },
      "source": [
        "train_story_text[0]"
      ],
      "execution_count": 488,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "['Mary',\n",
              " 'moved',\n",
              " 'to',\n",
              " 'the',\n",
              " 'bathroom',\n",
              " '.',\n",
              " 'Sandra',\n",
              " 'journeyed',\n",
              " 'to',\n",
              " 'the',\n",
              " 'bedroom',\n",
              " '.']"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 488
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "igLOVJ9hsNji",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "fd497b8f-e2a5-4ccb-c765-35dbb176fef4"
      },
      "source": [
        "train_story_seq[0]"
      ],
      "execution_count": 489,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[23, 1, 22, 31, 2, 29, 19, 27, 22, 31, 12, 29]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 489
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "9eVZhTiHsdQS",
        "colab_type": "text"
      },
      "source": [
        "## Building the Model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "89p9iRVosbEZ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Imports\n",
        "from keras.models import Sequential, Model\n",
        "from keras.layers.embeddings import Embedding\n",
        "from keras.layers import Input, Activation, Dense, Permute, Dropout, add, dot, concatenate, LSTM"
      ],
      "execution_count": 490,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8LKS-VZmsjw4",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# We need to create the placeholders \n",
        "#The Input function is used to create a keras tensor\n",
        "#PLACEHOLDER shape = (max_story_len,batch_size)\n",
        "#These are our placeholder for the inputs, ready to recieve batches of the stories and the questions\n",
        "input_sequence = Input((max_story_len,)) #As we dont know batch size yet\n",
        "question = Input((max_question_len,))"
      ],
      "execution_count": 491,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Xci-Je3ysnpX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Create input encoder M:\n",
        "input_encoder_m = Sequential()\n",
        "input_encoder_m.add(Embedding(input_dim=vocab_len,output_dim = 64)) #From paper\n",
        "input_encoder_m.add(Dropout(0.3))\n",
        "\n",
        "#Outputs: (Samples, story_maxlen,embedding_dim) -- Gives a list of the lenght of the samples where each item has the\n",
        "#lenght of the max story lenght and every word is embedded in the embbeding dimension"
      ],
      "execution_count": 492,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5bVFkB7jsqtN",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Create input encoder C:\n",
        "input_encoder_c = Sequential()\n",
        "input_encoder_c.add(Embedding(input_dim=vocab_len,output_dim = max_question_len)) #From paper\n",
        "input_encoder_c.add(Dropout(0.3))\n",
        "\n",
        "#Outputs: (samples, story_maxlen, max_question_len)"
      ],
      "execution_count": 493,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "2x1zxmUAss_y",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Create question encoder:\n",
        "question_encoder = Sequential()\n",
        "question_encoder.add(Embedding(input_dim=vocab_len,output_dim = 64,input_length=max_question_len)) #From paper\n",
        "question_encoder.add(Dropout(0.3))\n",
        "\n",
        "#Outputs: (samples, question_maxlen, embedding_dim)"
      ],
      "execution_count": 494,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tE4-QCnpsv5O",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Now lets encode the sequences, passing the placeholders into our encoders:\n",
        "input_encoded_m = input_encoder_m(input_sequence)\n",
        "input_encoded_c = input_encoder_c(input_sequence)\n",
        "question_encoded = question_encoder(question)"
      ],
      "execution_count": 495,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "q6swzTKOs50p",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Use dot product to compute similarity between input encoded m and question \n",
        "#Like in the paper:\n",
        "match = dot([input_encoded_m,question_encoded], axes = (2,2))\n",
        "match = Activation('softmax')(match)"
      ],
      "execution_count": 496,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nrZjOehbs_My",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#For the response we want to add this match with the ouput of input_encoded_c\n",
        "response = add([match,input_encoded_c])\n",
        "response = Permute((2,1))(response) #Permute Layer: permutes dimensions of input"
      ],
      "execution_count": 497,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "C8jxEupntBz-",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Once we have the response we can concatenate it with the question encoded:\n",
        "answer = concatenate([response, question_encoded])"
      ],
      "execution_count": 498,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "b0O83taUtFe_",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "4ca16ac1-3a5c-4f4b-9e79-e2c9fb83cb0c"
      },
      "source": [
        "\n",
        "answer"
      ],
      "execution_count": 499,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<tf.Tensor 'concatenate_7/concat:0' shape=(None, 6, 220) dtype=float32>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 499
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Y8WBL0m_tp-f",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Reduce the answer tensor with a RNN (LSTM)\n",
        "answer = LSTM(32)(answer)"
      ],
      "execution_count": 500,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c0gh_i0sttFP",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Regularization with dropout:\n",
        "answer = Dropout(0.5)(answer)\n",
        "#Output layer:\n",
        "answer = Dense(vocab_len)(answer) #Output shape: (Samples, Vocab_size) #Yes or no and all 0s"
      ],
      "execution_count": 501,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vJaw7j05tycV",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Now we need to output a probability distribution for the vocab, using softmax:\n",
        "answer = Activation('softmax')(answer)"
      ],
      "execution_count": 502,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "aSAYyN_Ht3Vk",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Now we build the final model:\n",
        "model = Model([input_sequence,question], answer)"
      ],
      "execution_count": 503,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "KTqh7YZ4t6oL",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "model.compile(optimizer='adam', loss = 'categorical_crossentropy', metrics = ['accuracy'])\n",
        "#Categorical instead of binary cross entropy as because of the way we are training\n",
        "#we could actually see any of the words from the vocab as output\n",
        "#however, we should only see yes or no"
      ],
      "execution_count": 504,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "td9chEqHt-Xk",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 693
        },
        "outputId": "e522f30c-8a46-4525-bc68-ad9b944392ab"
      },
      "source": [
        "model.summary()"
      ],
      "execution_count": 505,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model_9\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input_13 (InputLayer)           (None, 156)          0                                            \n",
            "__________________________________________________________________________________________________\n",
            "input_14 (InputLayer)           (None, 6)            0                                            \n",
            "__________________________________________________________________________________________________\n",
            "sequential_19 (Sequential)      multiple             2432        input_13[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "sequential_21 (Sequential)      (None, 6, 64)        2432        input_14[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "dot_7 (Dot)                     (None, 156, 6)       0           sequential_19[1][0]              \n",
            "                                                                 sequential_21[1][0]              \n",
            "__________________________________________________________________________________________________\n",
            "activation_13 (Activation)      (None, 156, 6)       0           dot_7[0][0]                      \n",
            "__________________________________________________________________________________________________\n",
            "sequential_20 (Sequential)      multiple             228         input_13[0][0]                   \n",
            "__________________________________________________________________________________________________\n",
            "add_7 (Add)                     (None, 156, 6)       0           activation_13[0][0]              \n",
            "                                                                 sequential_20[1][0]              \n",
            "__________________________________________________________________________________________________\n",
            "permute_7 (Permute)             (None, 6, 156)       0           add_7[0][0]                      \n",
            "__________________________________________________________________________________________________\n",
            "concatenate_7 (Concatenate)     (None, 6, 220)       0           permute_7[0][0]                  \n",
            "                                                                 sequential_21[1][0]              \n",
            "__________________________________________________________________________________________________\n",
            "lstm_7 (LSTM)                   (None, 32)           32384       concatenate_7[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "dropout_28 (Dropout)            (None, 32)           0           lstm_7[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "dense_7 (Dense)                 (None, 38)           1254        dropout_28[0][0]                 \n",
            "__________________________________________________________________________________________________\n",
            "activation_14 (Activation)      (None, 38)           0           dense_7[0][0]                    \n",
            "==================================================================================================\n",
            "Total params: 38,730\n",
            "Trainable params: 38,730\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZRgC-YPuuBLP",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\n",
        "history = model.fit([inputs_train,questions_train],answers_train, batch_size = 32, epochs = 100, validation_data = ([inputs_test,questions_test],answers_test))"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tvQm1_iMuU0X",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "filename = 'chatbot_model.h10'\n",
        "model.save(filename)"
      ],
      "execution_count": 507,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "owhiWdjIvCV7",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 747
        },
        "outputId": "5467b570-4da6-4c98-86a3-54196872170e"
      },
      "source": [
        "#Lets plot the increase of accuracy as we increase the number of training epochs\n",
        "#We can see that without any training the acc is about 50%, random guessing\n",
        "import matplotlib.pyplot as plt\n",
        "%matplotlib inline\n",
        "print(history.history.keys())\n",
        "# summarize history for accuracy\n",
        "plt.figure(figsize=(12,12))\n",
        "plt.plot(history.history['accuracy'])\n",
        "plt.plot(history.history['val_accuracy'])\n",
        "plt.title('model accuracy')\n",
        "plt.ylabel('accuracy')\n",
        "plt.xlabel('epoch')\n",
        "plt.legend(['train', 'test'], loc='upper left')\n",
        "plt.show()\n"
      ],
      "execution_count": 508,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "dict_keys(['val_loss', 'val_accuracy', 'loss', 'accuracy'])\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 864x864 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": [],
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Tkabej_bNIj4",
        "colab_type": "text"
      },
      "source": [
        "### Test the Model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FWDKkrQivG36",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 72
        },
        "outputId": "bee36cdd-a71b-4930-92dd-21f64537a09b"
      },
      "source": [
        "#To load a model that we have already trained and saved:\n",
        "from keras.models import load_model\n",
        "model=load_model('/content/chatbot_model.h10')\n",
        "\n"
      ],
      "execution_count": 544,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/indexed_slices.py:434: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
            "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "48NixxeuvNFh",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Lets check out the predictions on the test set:\n",
        "#These are just probabilities for every single word on the vocab\n",
        "pred_results = model.predict(([inputs_test,questions_test]))"
      ],
      "execution_count": 545,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3HUAxjFTvPrt",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 260
        },
        "outputId": "d69030dd-c21a-42a5-bfc7-f503c9824bf7"
      },
      "source": [
        "#First test data point\n",
        "test_data[0]"
      ],
      "execution_count": 546,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(['Mary',\n",
              "  'got',\n",
              "  'the',\n",
              "  'milk',\n",
              "  'there',\n",
              "  '.',\n",
              "  'John',\n",
              "  'moved',\n",
              "  'to',\n",
              "  'the',\n",
              "  'bedroom',\n",
              "  '.'],\n",
              " ['Is', 'John', 'in', 'the', 'kitchen', '?'],\n",
              " 'no')"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 546
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ibR96EZ5vTtW",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 191
        },
        "outputId": "137286e0-0311-4856-fe44-7e0aa1778a2a"
      },
      "source": [
        "#These are the probabilities for the vocab words using the 1st sentence\n",
        "pred_results[0]"
      ],
      "execution_count": 547,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "array([7.8486398e-14, 7.4002355e-14, 5.8381902e-14, 8.4658056e-14,\n",
              "       6.0745449e-14, 6.4910581e-14, 5.6233847e-14, 6.4374863e-14,\n",
              "       3.1387189e-04, 5.4258000e-14, 7.3613817e-14, 5.4822547e-14,\n",
              "       5.5996244e-14, 4.9346811e-14, 9.7374318e-14, 5.3766385e-14,\n",
              "       1.0897339e-13, 6.1730894e-14, 6.3984977e-14, 1.0048812e-13,\n",
              "       9.2711822e-14, 1.1036671e-13, 9.8954543e-14, 5.6103790e-14,\n",
              "       6.7933777e-14, 5.6448556e-14, 9.9968612e-01, 7.4600265e-14,\n",
              "       7.1462001e-14, 6.4112147e-14, 5.9549283e-14, 5.2002256e-14,\n",
              "       6.2774323e-14, 7.8156990e-14, 6.6290635e-14, 5.7822148e-14,\n",
              "       9.0104295e-14, 7.1634220e-14], dtype=float32)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 547
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "5mQvpgAPvXOL",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "val_max = np.argmax(pred_results[0])"
      ],
      "execution_count": 548,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sieE2YhYvhlc",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "2cc266bb-9fed-4940-e55a-3d91cb0e7887"
      },
      "source": [
        "for key,val in tokenizer.word_index.items():\n",
        "    if val == val_max:\n",
        "        k = key\n",
        "print(k)"
      ],
      "execution_count": 549,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "no\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oO1AHa3QvqYh",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "ebce41a4-b14f-4329-ae38-847b1a069abe"
      },
      "source": [
        "#See probability:\n",
        "pred_results[0][val_max]"
      ],
      "execution_count": 550,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.9996861"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 550
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oLWAj8Iv9hnU",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 72
        },
        "outputId": "ae3ad6a5-2b69-486b-def0-8bfcfc65c1ad"
      },
      "source": [
        "#To load a model that we have already trained and saved:\n",
        "from keras.models import load_model\n",
        "new_model=load_model('/content/chatbot_model.h10')"
      ],
      "execution_count": 509,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/indexed_slices.py:434: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.\n",
            "  \"Converting sparse IndexedSlices to a dense Tensor of unknown shape. \"\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "W-P40kXHvtGV",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 659
        },
        "outputId": "ccc3cc84-53ad-49d6-df53-124335ff3ea0"
      },
      "source": [
        "\n",
        "#Now, we can make our own questions using the vocabulary we have\n",
        "vocab"
      ],
      "execution_count": 510,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "{'.',\n",
              " '?',\n",
              " 'Daniel',\n",
              " 'Is',\n",
              " 'John',\n",
              " 'Mary',\n",
              " 'Sandra',\n",
              " 'apple',\n",
              " 'back',\n",
              " 'bathroom',\n",
              " 'bedroom',\n",
              " 'discarded',\n",
              " 'down',\n",
              " 'dropped',\n",
              " 'football',\n",
              " 'garden',\n",
              " 'got',\n",
              " 'grabbed',\n",
              " 'hallway',\n",
              " 'in',\n",
              " 'journeyed',\n",
              " 'kitchen',\n",
              " 'left',\n",
              " 'milk',\n",
              " 'moved',\n",
              " 'no',\n",
              " 'office',\n",
              " 'picked',\n",
              " 'put',\n",
              " 'the',\n",
              " 'there',\n",
              " 'to',\n",
              " 'took',\n",
              " 'travelled',\n",
              " 'up',\n",
              " 'went',\n",
              " 'yes'}"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 510
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9ktOaJW5vyFm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "my_story = \"John left the kitchen . Sandra dropped the football in the garden .\"\n",
        "my_question = \"Is John left the kitchen ?\""
      ],
      "execution_count": 537,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rW5QXcqiwDU4",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Put the data in the same format as before\n",
        "my_data = [(my_story.split(), my_question.split(),'yes')]"
      ],
      "execution_count": 538,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0CPCpQJTwGA8",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "\n",
        "#Vectorize this data\n",
        "my_story, my_ques, my_ans = vectorize_stories(my_data)"
      ],
      "execution_count": 539,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NPK0L_M8wJWQ",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 156
        },
        "outputId": "8d54744f-5b33-42af-98da-c2443277a0e2"
      },
      "source": [
        "\n",
        "#Make the prediction\n",
        "pred_results = new_model.predict(([my_story,my_ques]))\n",
        "print(pred_results)"
      ],
      "execution_count": 540,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[3.5882118e-12 3.2988007e-12 2.6436980e-12 4.2214197e-12 2.5450397e-12\n",
            "  3.3433373e-12 2.5426766e-12 2.6117769e-12 7.6728815e-01 2.5145446e-12\n",
            "  3.2566022e-12 2.5355139e-12 2.6344311e-12 2.0178902e-12 5.0790839e-12\n",
            "  2.1226146e-12 4.8332770e-12 2.6149920e-12 3.0280969e-12 4.6232180e-12\n",
            "  4.3193274e-12 5.6177571e-12 4.4199024e-12 2.4795582e-12 3.0544727e-12\n",
            "  2.5143386e-12 2.3271182e-01 4.0629765e-12 3.5000338e-12 3.0308243e-12\n",
            "  3.1051019e-12 2.3959335e-12 2.7926478e-12 3.9747745e-12 3.0705009e-12\n",
            "  2.8089467e-12 5.0486017e-12 3.5202663e-12]]\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "U62CxlxfwMko",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "val_max = np.argmax(pred_results)"
      ],
      "execution_count": 541,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-eeAzE8BwQyR",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "d07c5745-af73-4563-9e42-10f62b215fd9"
      },
      "source": [
        "#Correct prediction!\n",
        "for key,val in tokenizer.word_index.items():\n",
        "    if val == val_max:\n",
        "        k = key\n",
        "print(k)"
      ],
      "execution_count": 542,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "yes\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "YzxYZAR3wTJZ",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "b3735637-6e6a-46b5-d298-c12f482f7fab"
      },
      "source": [
        "\n",
        "#Confidence\n",
        "pred_results[0][val_max]\n",
        "\n",
        "\n"
      ],
      "execution_count": 543,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.76728815"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 543
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xJa10DFXBI1s",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 311,
      "outputs": []
    }
  ]
}